Derivative-enhanced Deep Operator Network

Authors: Yuan Qiu, Nolan Bridges, Peng Chen

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments validate the effectiveness of our approach.
Researcher Affiliation Academia Yuan Qiu, Nolan Bridges, Peng Chen Georgia Institute of Technology, Atlanta, GA 30332 {yuan.qiu, bridges, pchen402}@gatech.edu
Pseudocode No The paper does not contain pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes The code for data generation, model training and inference, as well as configurations to reproduce the results in this paper can be found at https://github.com/qy849/DE-Deep ONet.
Open Datasets No We generate Ntrain = 1500 and Ntest = 500 input-output pairs (m(i), u(i)) for training and testing, respectively.
Dataset Splits No The paper specifies training and test sets but does not explicitly mention a separate validation set split.
Hardware Specification Yes Table 4: Wall clock time (seconds/iteration with batch size 8) for training on a single NVIDIA RTX A6000 GPU; Table 2: Wall clock time (in seconds) for data generation on 2 AMD EPYC 7543 32-Core Processors
Software Dependencies No The paper mentions software like FEniCS [31] and hIPPYlib [28], as well as torch.func.jacrev and torch.vmap (implying PyTorch), but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We train each model for 32768 iterations (with the same batch size 8) using an Adam W optimizer [34] and a Step LR learning rate scheduler (We disable learning rate scheduler for DE-Deep ONet).